Meta's Muse Image: the new AI model that can pull Instagram users into generated photos

Meta has released a new artificial-intelligence image generator called Muse Image, and within hours of the announcement users were already objecting to one of its capabilities. The model, reported by The Verge, is the first image-generation system built by Meta's Superintelligence Labs, the division the company assembled to pursue its most ambitious AI work, and it now powers the image tools across Meta's apps.

Muse Image sits at the centre of Meta's consumer AI strategy. According to the company's announcement, it powers image-making tools in the Meta AI app, Instagram and WhatsApp, and is coming to Facebook and Messenger. It is part of a growing family of Muse models that Meta says are replacing its earlier Llama lineup as the engines behind its user-facing AI features.

The feature that drew immediate criticism is the ability to pull other Instagram users into AI-generated photos. In practice, that means the system can incorporate a likeness of someone other than the person creating the image, a capability that sits directly on top of the platform's most sensitive question: control over one's own image. Many users reacted with concern about being inserted into pictures they did not create or consent to.

Meta describes the model in technical terms that signal where the industry is heading. Alexandr Wang, who the company hired to lead Superintelligence Labs, said on Threads that Muse Image is "agentic", meaning it works with a companion large language model called Muse Spark to "reason through your prompt, search the web, and plan before it generates". In other words, the system is designed to interpret a request and take steps, rather than produce a single image in one pass.

That agentic framing matters because it marks a shift in how image generators work. Earlier tools mapped a text prompt more or less directly to an image. A system that reasons, searches and plans can produce more elaborate results, but it also makes more decisions on the user's behalf, which raises the stakes when those decisions involve real people's likenesses.

The backlash reflects a tension Meta has faced repeatedly. Its platforms hold enormous quantities of personal photos, and integrating generative AI into them promises powerful creative features while inviting exactly the privacy fears users voiced here. The ability to place a recognisable person into a synthetic scene is precisely the kind of function that can be delightful in the right hands and troubling in the wrong ones.

Context makes the concern sharper. AI-generated imagery has already fuelled worries about deepfakes, non-consensual images and misinformation, and a feature that lowers the barrier to putting a specific individual into a fabricated photo touches all of those anxieties. When such a tool is embedded directly into apps used by billions, the scale of potential misuse grows accordingly.

Meta has generally responded to these concerns with a mix of controls and policies, such as labelling AI-generated content and offering settings that govern how a person's images can be used. How robust those safeguards are for Muse Image specifically, and how easily they can be overridden, will shape whether the backlash fades or hardens as more users encounter the feature.

The launch also underlines the competitive stakes. Meta is racing rivals to define what consumer AI looks like, and image generation woven into the world's largest social platforms is a powerful position. Building its own frontier models through Superintelligence Labs, rather than relying on outside systems, is part of a broader industry pattern of large technology companies investing to control the core AI that powers their products.

For users, the practical takeaway is to check the settings. As generative tools become default features rather than opt-in experiments, understanding what a platform can do with your photos and likeness, and what controls exist to limit it, becomes part of ordinary digital literacy. Muse Image is a capable new tool, and the early reaction is a reminder that capability and consent do not automatically move in step.